Search Analysis for the Underwater Wreckage of Flight 447

Lawrence D. Stone Johan P. Strumpfer Colleen M. Keller Partners for Change Thomas M. Kratzke Clovelly, South Africa Metron Inc [email protected] Reston Virginia, U.S.A.

[email protected] [email protected] [email protected]

On 1 June 2009 Air France Flight 447, with 228 proved unsuccessful. The Phase III search, which took passengers and crew aboard, disappeared over the place from 2 April to 24 May 2010, consisted of South Atlantic during a night flight from additional side looking searches using REMUS to . An international air and surface search effort AUVs operated by the Woods Hole Oceanographic located the first floating debris during the sixth day of Institute (WHOI) and using the ORION towed side- search. Three phases of unsuccessful search for the looking sonar operated by the US Navy with assistance underwater wreckage ensued. Phase I was a passive from Phoenix International. The search also used a Triton acoustic search for the aircraft’s underwater locator ROV provided by Seabed AS (Norway). It was likewise beacons. Phases II and III were side-looking sonar unsuccessful. searches scanning the ocean bottom for the wreckage In July of 2010, Metron was tasked by the BEA to field. In July of 2010 the French Bureau d’Enquêtes et review the search and to produce an updated posterior d’Analyses tasked Metron to review the searches and probability map for the location of the underwater produce posterior probability maps for the location of wreckage. To accomplish this Metron reviewed and the wreckage. These maps were used to plan the next modified the distributions developed in 2009. A new phase of search beginning in March 2011. On April 3, prior was developed based on studies by the BEA, the after one week of search, the wreckage was located in a Russian Interstate Aviation Group (MAK), and a new high probability area of the map. reverse drift analysis using updated current estimates. Metron analyzed the effectiveness of Phase III side Keywords: Search, Bayesian, Posterior Distribution. looking sonar searches performed by the WHOI REMUS and the US Navy ORION sensors and computed an 1 Introduction updated posterior probability distribution for the location On 1 June 2009 Air France Flight 447, an Airbus 330-200 of the wreckage using the new prior distribution and with 228 passengers and crew, disappeared over the South incorporating the unsuccessful phase I and II searches Atlantic during a night flight from Rio de Janeiro to Paris. performed during 2009, as well as the unsuccessful Phase An international air and surface search effort recovered III searches performed by REMUS and ORION in 2010 the first wreckage on June 6th, five and one half days after as well as the Triton ROV searches. Metron also the accident. More than 1000 pieces of the aircraft and 50 accounted for the unsuccessful surface searches bodies were recovered and their positions logged. A performed by aircraft and ships between 1 and 6 June French submarine as well as French and American 2009. research teams searched acoustically for the Underwater Metron’s previous work in search applications, Locator Beacons (ULBs, or “pingers”) on each of the two detailed in references [1,2,3], includes searches for the ’s “black boxes” for 31 days from 10 June U.S. nuclear submarine Scorpion, the SS Central America, to 10 July 2009 with no results. and Steve Fossett’s crash site. In addition, Metron played In early July of 2009 the French Bureau d’Enquêtes a key role in developing the U. S Coast Guard’s Search et d’Analyses pour la sécurité de l’aviation civile, and Rescue Optimal Planning System (SAROPS) which abbreviated as BEA, contacted Metron for assistance in has been successfully employed to plan and execute preparing for Phase II of the search, utilizing side-looking searches for ships and personnel lost at sea [4]. This sonar to scan the ocean bottom for the wreckage field. paper describes the results of Metron’s Air France 447 The Phase II side looking sonar search, performed (AF 447) analysis. by the Pourquoi Pas? from 27 July to 17 August 2009, 2 Approach estimates CDP at the conclusion of each underwater search phase. Our approach to this search planning problem is rooted The following chart summarizes the various steps of in classical Bayesian inference, which allows the this approach and also references other figures that are organization of available data with associated used throughout this report. In Phase I, the Reverse Drift uncertainties and computation of the Probability and Surface Search blocks, shown in green, use the Distribution Function (PDF) for target location given SAROPS environmental module that simulates winds and these data. In following this approach, the first step was currents in the search zone. The other Phase I blocks are to gather the available information about the location of based on flight dynamics computations and a study the impact site of the aircraft. Using a Bayesian approach undertaken on loss of control accidents during flight. we organized this material into consistent scenarios, quantified the uncertainties with probability distributions, Fig 13 weighted the relative likelihood of each scenario, and performed a simulation to produce a prior PDF for the Fig 11 location of the wreck. This is the same methodology that was pioneered in [1] and incorporated into SAROPS. Fig 9 Next we estimated the effect of the unsuccessful search efforts. These efforts included air and surface searches Fig 7 for floating debris and underwater searches in Phases I, II, Fig 6 and III. The goal of the Phase I search was to detect signals from the flight recorders’ ULBs. The Phase II and Fig 5 III searches involved the use of side-looking sonar and cameras to detect the underwater debris field of the wreck Fig 3 Fig 4 of the AF 447 flight. For each search, we enlisted sensor experts and knowledge of the sea state, visibility, underwater geography, and water column conditions to estimate sensor performance. The results of the search Figure 1. Summary of PDF Computation assessment, combined mathematically with the prior PDF of the impact site, yielded the posterior PDF for the As an excursion, we computed the posterior PDF impact location given the unsuccessful search efforts. assuming that both ULB “pingers” failed to function Posterior PDFs after each phase of the search are during the Phase I search. On 3 April 2011, the wreckage presented in Section 4 along with the estimated was found in a high probability area of this distribution effectiveness of the search in terms of Cumulative (Figure 14). From the location of the wreckage, it appears Detection Probability (CDP). that the ULBs either failed to operate properly or failed to The posterior distribution given in Section 4 provided be detected, which greatly increased the difficulty of the guidance for the location of the wreckage and the amount search. In early May the Flight Data Recorder and of additional search effort required to obtain a high Voice Recorder were recovered. probability of success. Section 3 of this paper describes the method for 3 Prior PDF for Impact Location producing the prior (to the surface search) PDF for impact In this section we compute the prior (before surface location. This distribution is composed of two search) PDF for impact location. This PDF has two components. The first component, called the Flight components, a flight dynamics and a reverse drift Dynamics (FD) prior, is based on flight dynamics component. considerations and information from past crashes. The second component of this prior is derived from the 3.1 Flight Dynamics Prior information provided by the detection and recovery of floating debris from the wreckage of the aircraft on 6 June This prior is the mixture of two distributions. The first – 10 June. This information was used to produce a is based on purely flight dynamics considerations about Reverse Drift (RD) prior. The FD and RD priors were the maximum distance the aircraft could have feasibly blended to produce a surface search prior. traveled from the time of its last reported position (last In section 4, the effect of the unsuccessful surface known position (LKP)) at 35,000 ft. altitude to the time searches during 1 June – 6 June 2009 conducted by when a scheduled response from the Aircraft aircraft and ships was used to compute the surface search Communications Addressing and Reporting System posterior. This posterior became the prior for the passive (ACARS) was not received. ACARS is a maintenance and side-scan sonar searches in Phases I, II, and III. The and logistics reporting system that sends out position remainder of section 4 computes the posterior PDF and reports based on GPS roughly every 10 minutes. The impact time was estimated based on the time of the last ACARS message received and the expectation (unfulfilled) of a subsequent message in the next 60 3.2 Reverse Drift Prior seconds. The end of the flight occurred between 2 h 14 min 26 sec and 2 h 15 min 14 sec - see page 39 of [5]. An The reverse drift (RD) prior was computed using analysis was performed by the BEA and reported in SAROPS and data on currents and winds to reverse the reference [6] which produced a uniform distribution over motion of recovered floating debris pieces back to the the disk of radius 40 NM centered at the LKP. This is the time of impact. first distribution. The U. S. Coast Guard employs SAROPS for all their The second distribution is based on data from nine planning. SAROPS allows a search commercial aircraft accidents involving loss of control. planner to define scenarios, obtain the winds and currents This analysis was performed by the Russian Interstate necessary to compute drift trajectories, estimate effective Aviation Group [7] and the BEA. Figure 2 shows the sweep widths for search sensors, develop probability cumulative distribution of distance (pro-rated to 35,000 ft. distributions for search object location, and find near altitude) flown from the beginning of the emergency optimal search plans given the amount of search effort situation to impact of the aircraft. available.

Fraction of Impact Locations within Distance D of Beginning of Emergency In order to compute an RD scenario, one must have an 1 estimate of the surface currents in the area of the crash 0.9 during 1 – 10 June 2009. The BEA commissioned a 0.8 group of oceanographic experts to review the data 0.7 available for estimating these currents. The results are 0.6 reported in the Drift Group Report [8]. Because the area

0.5 is near the equator and in the middle of the Atlantic, the Fraction

0.4 currents are complex and difficult to estimate. In addition

0.3 the remote nature of the crash site meant that there were few meteorological measurements to provide a basis for 0.2 current estimates. Because of the complexity of the 0.1 currents and the lack of data, there is substantial 0 0 2 4 6 8 10 12 14 16 18 20 uncertainty in these estimates. D (NM) Figure 2. Cumulative Distribution of Distance Traveled from Metron used the ANALYSE_75KM_LPO current Beginning of Emergency to Impact Location estimates to compute a reverse drift prior. These The analysis shows that all impact points are contained estimates were produced as a result of the work of the within a 20-NM radius circle from the point at which the Drift Group. However, we have given the results based emergency situation began. These results are represented on these estimates a low weight (30%) in producing the by a circular normal distribution with center at the LKP prior and posterior distributions because of the great and standard deviation 8 NM along both axes. For the uncertainty associated with the estimates. Flight Dynamics (FD) prior, we chose a mixture weighted To produce the RD Prior Metron used the positions and by 50% for the uniform over 40 NM distribution and 50% recovery times of the 33 bodies that were located from 6 – for the circular normal distribution truncated at 40 NM 10 June 2009 [5]. Some bodies were recovered in groups. from the LKP. This distribution is shown in Figure 3 The positions of the bodies or groups of bodies were drifted back in time using the ANALYSE_75KM_LPO current estimates. There are two components of drift. Drift due to ocean current and drift due to wind which is called leeway. We did not apply reverse drift to pieces of debris that were recovered during this time because we do not have good models for the effect of leeway on this type of debris. Data obtained in September 2009 from experiments on the drift of a manikin modified to simulate a deceased person floating in the water (PIW) is reported in [9]. From these experiments, the authors produced the leeway model shown by equation (1) below. This model, which is based on empirical data, produces a total drift speed of roughly 2.35% of wind speed and includes a cross wind component. The equations in (1) were used in SAROPS to account for the leeway of the bodies. Figure 3. Flight Dynamics Prior DW=+1.17 10.2cm/s WL 10m (1) CWL =+ 0.04W10m 3.9cm/s where W10m is in m/s and DWL and C WL are in cm/s. the search efforts in Phases I – III, and combine them to Figure 4 shows the reverse drift distribution produced compute the posterior PDF on impact location given in this fashion and truncated at the 40 NM circle. failure of these efforts. The unsuccessful searches considered in this analysis include the ones listed below. Unsuccessful Surface Searches: 1 June to 5 June 2009. • The air and ship search efforts failed to positively identify and recover floating debris or bodies during the period from 1 June to 5 June. The first piece of debris was recovered and identified on June 6th. Phase I: 10 June to 10 July 2009 • Passive acoustic searches for the “black box” Underwater Locator Beacons (ULBs) by the US Navy Towed Pinger Locators (TPLs). • Search by the IFREMER Victor Remotely Operated Vehicle (ROV) Phase II: 27 July to 17August 2009 • Side looking sonar search by the IFREMER deep sonar towed by the Pourquoi Pas? Figure 4. Reverse Drift Prior Phase III: 2 April – 24 May 2010 • Side-scan sonar search by three REMUS 3.3 Prior Before Surface Search Autonomous Underwater Vehicles (AUVs) and visual/sonar search by the Triton ROV. The prior distribution before surface search by aircraft and ships is a mixture of 70% of the FD Prior given in • Search by the USN Orion towed side-scan sonar section 3.1 and 30% of the RD Prior given in section 3.2. system. The resulting distribution is shown in Figure 5. 4.1 Accounting for Unsuccessful Search The SAROPS program uses a large number N of simulated points or particles to represent the probability distribution on the path or location of a search object. th The n particle has weight wn for nN=1,… , . Initially all weights are set equal. The weight is the probability that the particle represents the search object’s location or path. The SAROPS PDF in Figure 5 was produced by adding the weights (probabilities) of the particles in each cell to obtain the probability that the impact point is in that cell. These probabilities are represented by the color code progressing from red to yellow to blue as cells progress from high to low probability. The particles form the actual distribution computed by SAROPS. The cells are used simply as a method of display. If an unsuccessful search takes place, we compute the 1 Figure 5. PDF for Impact Location Prior to Surface Search probability pnd () that the search would have detected the search object if it were located where particle n is for 4 Posterior Distribution Given nN=1,… , . From this we compute the posterior distribution on object location using Bayes’ rule as Unsuccessful Search follows). Effort that fails to find the search object provides 1()− pnw1 1 ()dn (negative) information about the object’s location. This wn = N for nN=1,… , (2) 1()− pn1 ′ w information is incorporated into the posterior distribution ∑ n′=1()d n′ on impact location through the use of Bayes’ rule in the 1 fashion described in Section 4.1 In this section we where wn is the posterior probability that particle n estimate the effectiveness of the surface search effort and represents the object’s location. We can see from (2) that 1 calculated a new PDF for the impact point. The result of if pnd () is close to 1, the posterior probability on particle n will tend to be low. Correspondingly those particles this calculation is shown in Figure 6. 1 with low values of pnd () will tend to have high posterior probabilities. If the particles are moving and the search sensor is moving, SAROPS accounts for both of these motions in calculating pnd () for each particle. 4.2 Aircraft and Ship Surface Search Searches for debris by Brazilian and French aircraft were conducted from June 1st to June 26th 2009. These searches were unsuccessful until June 6 when debris and bodies from the aircraft were first recovered. Analysis of the unsuccessful air searches and ship searches prior to 6 June provides negative search information that we use to decrease the probability on some particles and increase it on others according to Bayes’ rule. Search paths for each sortie were entered into SAROPS, Figure 6. Surface Search Posterior PDF along with information on altitude, speed, aircraft type, Underwater Search Prior. For the purpose of meteorological visibility, sea state, and expected sensor evaluating the underwater search, we formed a prior performance against a raft-sized target (specifically a which is a mixture of 30% of the Surface Search Posterior four-man raft), which was taken as a surrogate for the in Figure 6 and 70% of the FD Prior in Figure 2. This detectability of a large piece of debris such as the galley. PDF is shown in Figure 7. We have given the Surface In addition to the air search, we included the effort of the Search Posterior a low weight in this PDF because it ship Douce France that searched in the vicinity of LKP depends heavily on estimates of currents in the area of the on June 1st. crash during 1 – 10 June. For the reasons discussed in Two Embraer R-99 and a French Section 3.2, we have low confidence in these estimates Air Force E-3F (AWACS) were also involved in the search. They patrolled at high altitude and used their airborne radars to search for possible reflections from the ocean surface. The (vast) surfaces that they covered are not accounted for in the search analysis. Only the low altitude visual searches in the vicinity of the 40 NM circle are included. To evaluate the effect of the unsuccessful surface search, we started with the surface search prior distribution given in Figure 5. We allocated simulation particles in accordance with the probability density in each cell. We then used SAROPS to “drift” these particles forward in time from the time of impact though 6 June. In the process of doing this we accounted for the unsuccessful aircraft and ship search during that time. The search object was assumed to have the drift and detection characteristics of a four-person life raft. If the particles were predicted to pass through an area searched Figure 7. Underwater Search Prior PDF by aircraft or the Douce France, their weight was appropriately reduced according to the estimated detection 4.3 Phase I Searches probability for that platform sortie as described in section In this section we describe the searches performed 4.1. The result is a set of particles whose weights during Phase I and compute the posterior PDFs resulting (probabilities) have been updated to incorporate the from these unsuccessful searches. unsuccessful air and ship search effort during those days. Acoustic Searches for the ULBs. The aircraft was Surface Search Posterior. We then pulled each equipped with two “black boxes”, the digital Flight Data particle back to its position at the time of impact keeping Recorder (FDR) and the Cockpit Voice Recorder (CVR). its weight the same as it was at the end of unsuccessful These ruggedized devices are designed to withstand the surface search. From these reweighted particles, we high impacts expected in a crash, and are fitted with an Underwater Locator Beacon (ULB) that activates when were likely to be working according to the manufacturer, contact is made with water. The batteries on the ULBs we discounted the TPL detection probability to 0.385. are certified to last at least 30 days. In the case of the ULBs fitted on this aircraft, the manufacturer stated that the duration of the transmission was of the order of forty days. The passive acoustic search for the FDR and CVR, which lasted 31 days and ended on 10 July 2009, primarily involved two tugs hired to assist in the search, Fairmount Glacier and Fairmount Expedition. The Fairmount ships’ search efforts overlaid the aircraft’s intended track. Both ships employed Towed Pinger Locators (TPLs) supplied by the US Navy. Personnel from Phoenix International operated the equipment. On-site tests indicated the equipment was functioning properly. The TPL sensors were assessed to detect the ULBs at a lateral range of 1730m with a POD of 0.90. This detection range accounted for the frequency of the ULBs’ emissions (37.5 kHz) and the assumed source level (160 Figure 8. Fairmount Glacier (orange) and Expedition dB). The TPLs were flying above the underwater terrain, (pink) TPL Search Tracks so we estimated that degradation due to terrain shadowing The posterior distribution after unsuccessful ULB was minimal. Environmental calculations showed that searches by the Fairmount ships is shown in Figure 9. deep water propagation in this area is basically direct path This is the posterior for the location of the underwater and the transmission loss and ambient noise are wreckage after the Phase I searches. The Cumulative sufficiently low to provide detection probabilities of 0.90 Detection Probability for the TPL searches is 0.41. or above, a number that we feel is conservative provided at least one of the ULBs was operating properly. The ships’ tracks, reconstructed from GPS data, are shown in Figure 8 below. The calculation of the probability of detection for the TPL search must account for the possibility that the ULBs were destroyed in the crash. Based on the condition of the wreckage recovered and after conferring with the BEA, we assumed a probability of 0.8 that a single ULB survived the crash If ULB survival is considered independent, then the probability of detecting at least one ULB given it is within lateral range 1730 m of the TPL as follows:

PD = Pr{} detect|both working Pr{} both working +Pr{} detect|one working Pr{} one working

22 =−()1 (0.1) (0.8) +()( 0.9 2(0.8)(0.2) )= 0.92 Figure 9. Posterior PDF after Phase I: CDP = 0.41 If the ULBs were mounted sufficiently close together to consider their chances of survival to be completely 4.4 Phase II Search dependent, then the probability of detecting at least one To continue the search after the pingers’ extinction, the ULB drops to 0.9×0.8 = 0.72. BEA decided to use the IFREMER side-looking sonar It is difficult to say whether the survival of the two towed by the Pourquoi Pas? which had the capability to ULBs should be considered independent events. In light complete the bathymetry survey of the area thanks to its of this uncertainty, it was decided to use a weighted hull-mounted multi-beam sonar. average of 0.25 for the independent and 0.75 for the Metron’s 2009 analysis of the Phase I search efforts dependent probabilities yielding a 0.77 detection was presented to the BEA during the planning for the probability given a ULB is within lateral range 1730 m of second phase of search. It was estimated that the a TPL search during the 30 day period from June 1 – June Pourquoi Pas? could cover three-to-four cells in the three 30 , 2009 which corresponds to the pingers rated 30 day weeks in which it would be on station at the accident site, lifetime. During the additional 10 days that the pingers achieving a 0.9 POD in those cells. The BEA chose to search the eastern half of J24, all of reviewed in accelerated playback and all items of interest K24 and L24, and the western half of M24 as shown in were further processed with sonar enhancement software. yellow in Figure 10. This constituted three full cells in a US Navy/Phoenix International operators/analysts row south/southeast of the LKP, 19 – 36 NM away at the evaluated the 1743 km2 as having been covered with the farthest point. Figure 11 shows the posterior PDF after highest degree of confidence. Two patches of bottom, the unsuccessful Phase II search. located in the peaks of some of the steepest slopes produced returns that were not interpretable. These were subsequently covered by REMUS 6000 AUVs. In the high confidence areas, we used a detection probability of 0.90. For the rest, we set the detection probability to 0.50 with the exception of the ridge areas which received 0.10.

Figure 10. Pourquoi Pas? Search Cells (yellow).

Figure 12. Search Areas for ORION (orange) and REMUS/Triton (grey) Within the searched area Phoenix International operators identified a number of small targets which BEA planed to investigate during search operations in 2011. Woods Hole Oceanographic Institution. The Seabed Worker deployed to the search area with three REMUS 6000 AUVs — two belonging to the Waitt Institute for Discovery and one to GEOMAR, and the Seabed Triton XLX 4000 ROV. The three REMUS vehicles covered the Figure 11. Posterior PDF after Phase II: CDP = 0.45 4,375 km2 area shown in grey in Figure 12. The REMUS side scan sonar maximum range is 600 – 700 m. Search 4.5 Phase III Searches legs were spaced a distance apart equal to the maximum There were two search efforts during Phase III. One range less 50 m which produced double coverage almost involved the US Navy/Phoenix International with assets everywhere within its search area. Most regions of steep onboard the Anne Candies and the other involved assets terrain, such as ridges, were imaged from both sides and operated by WHOI. Both efforts used side-scan sonar. were thus well searched. Exceptions were regions that The search area covered by these searches was, for the contained ridges that were not suitable for side-looking most part, determined by the search area sonar coverage. The Triton ROV was deployed in some recommendations made by the Drift Group in [8]. of these. US Navy/Phoenix International. The US We attributed a detection probability of 0.90 to the Navy/Phoenix International search was performed using areas shown in grey in Figure 12 with the exception of the the USN ORION towed side-scan sonar system which areas noted where we set the detection probability to 0.10. covered the 1900 square kilometer area of orange swaths 4.6 Posterior After Phase III Searches shown in Figure 12. The ORION was operated to cover a 2400 meter swath. Adjacent lines were spaced no greater Figure 13 shows the posterior PDF after the than 2000 meters apart. All records were monitored in unsuccessful searches from Phases I, II, and III. real time. During turns between track lines, the data were weighted scenario approach allowed inconsistent information to be combined with subjective weights that capture the confidence in each piece of data. The analysis of the detection effectiveness of each component of the search which produced the Bayesian posterior distributions shown in Figures 13 and 14 formed a solid basis for planning the next increment of search. In fact the phase IV search commenced in the center of the distribution and quickly found the wreckage. The success of this effort provides a powerful illustration of the value of this methodical, Bayesian approach to search planning. The full report [11] of this work is available on the BEA website. References Figure 13: Posterior PDF after Phase III: CDP = 0.58 [1] Henry R. Richardson and L. D. Stone, Operations Posterior Assuming Pingers Failed. If both pingers Analysis During the Underwater Search for Scorpion, failed to activate, the ULB search would have had no Naval Research Logistics Quarterly Vol 18 June 1971. chance of detecting the wreckage. If this were the case we would remove the ULB search in computing the posterior. [2] Lawrence D. Stone, Search for the SS Central The result of doing this is shown in Figure 14 below. America: Mathematical Treasure Hunting Interfaces Vol 22: January – February 1992. [3] Colleen M. Keller and M. A. Anderson, Applying Location of wreckage Optimal Search Theory to Inland SAR: Steve Fossett Case Study, October 2008. [4] U. S. Coast Guard Addendum to the United States National Search and Rescue Supplement to the IAMSAR Manual, (COMDTINST M16130.2E) September 2009. [5] Interim Report n°2 on the accident on 1st June 2009 to -203 registered F-GZCP operated by Air France, flight AF 447 Rio de Janeiro – Paris by Bureau d’Enquêtes et d’Analyses, 17 December 2009 (see http://www.bea.aero/en/enquetes/flight.af.447/flight.af.447.php) [6] AF447 Flight Path Analysis, BEA briefing, July 27,

2010 Figure 14. Posterior Assuming Pingers Failed: CDP = 0.29 [7] Presentation MAK AF447 Briefing, Russian 5 Conclusions Interstate Aviation Group, 16 October 2009 The posterior PDFs in Figures 13 and 14 provided good [8] Estimating the Wreckage Location of the Rio-Paris guidance for the phase IV search [10]. The wreckage is AF447, Report from Drift Group to BEA, June 30, 2010 located in an area thoroughly covered by the TPL passive [9] Ø. Breivik, A. Allen, C. Maisondieu, and J-C Roth acoustic search (see Figure 8) which suggests the locator Wind-induced drift of objects at sea: The leeway field beacons were not functioning properly. Figure 14 shows method Applied Ocean Research 2011 doi: the wreckage is located in a high probability area 10.1016/j.apor.2011.01.005. assuming the ULBs failed to operate properly. These results suggest is likely that either both ULBs failed or [10] Troadec,Jean-Paul. Undersea search operations to failed to be detected for some reason, and that this failure find the wreckage of the A 330, flight AF 447: the resulted in a long and difficult search for the wreckage. culmination of extensive searches. Note from BEA The BEA began recovery operations in April 2011. One Director, 8 April 2011, on the BEA website at of the goals is to recover the ULBs to determine if they http://www.bea.aero/en/enquetes/flight.af.447/note.from.bea.dire were functioning during the 30 day period after the crash. ctor.end.phase4.pdf Methodology. Our approach used careful and [11] Lawrence D. Stone, C M. Keller, T. M. Kratzke, and methodical consideration of all data available, with J. P. Strumpfer, Search Analysis for the Location of the associated uncertainties, to form an analytic assessment of AF447 Underwater Wreckage Report to BEA, 20 January the highest likelihood areas for future search efforts. The 2011.